4

Let $f:S\rightarrow \mathbb{R}^n$ be an isometry, where $S\subset \mathbb{R}^n$. Prove that there exist a matrix $A\in\mathbb{R}^{n\times n}$ such that $|\det A|=1$ and a vector $b\in \mathbb{R}^n$ such that $f(x)=Ax+b$ for all $x\in S$.

The second question I have is: What conditions must the set $S$ meet if the matrix $A$ and the vector $b$ are unique?

I suppose that the condition is: There exist $n+1$ points $s_1,\ldots,s_{n+1}\in S$ such that all of them don't lie in any hyperplane (i.e. $(n-1)$-dimensional affine subspace ). For example for $n=2$ there exist $s_1,s_2,s_3\in S$ which are vertices of a triangle. However, I don't know how to prove it.

Kulisty
  • 1,818

1 Answers1

5

Let $s_0\in S$. Let $U$ be the linear subspace of $\Bbb R^n$ spanned by al $s-s_0$, $s\in S$. Let $W$ be the linear subspace of $\Bbb R^n$ spanned by al $f(s)-f(s_0)$, $s\in S$. Then we can pick a basis $u_1,\ldots, u_m$ of $U$ such that all $u_i$ are of the form $s_i-s_0$, $s_i\in S$. If we let $w_i=f(s_i)-f(s_0)$ then the $w_i$ are also a basis of $W$. Clearly, $|w_i|=|f(s_i)-f(s_0)|=|s_i-s_0|=|u_i|$ and $|w_i-w_j|=|f(s_i)-f(s_j)|=|s_i-s_j|=|u_i-u_j|$. From this we conclude equality of scalar products, $w_i\cdot w_j=u_i\cdot u_j$, and hence that the linear map $\phi\colon U\to W$ given by $v_i\mapsto w_i$ is an isometric linear isomorphism.

Lemma. $f(x)=\phi(x-s_0)+f(s_0)$ for all $x\in S$.

Proof. First of all note that $x-s_0\in U$ and $f(x)-f(s_0)\in W$. Hence we can write $$ x=s_0+\sum a_i u_i,\qquad f(x)=f(s_0)+\sum b_i w_i.$$ The $a_i$ can be determined by a few length measurements inside $S$: We have for $1\le k\le m$ $$\tag1\begin{align}|x-s_k|^2&=|x-s_0|^2+|s_k-s_0|^2+2(s_k-s_0)(x-s_0)\\ &=|x-s_0|^2+u_k^2+2\sum_{i=1}^ma_iu_iu_k\end{align}$$ As the Gram matrix for the $u_i$ is invertible, we can compute the $a_i$ from $(1)$. Moreover, the Gram matrix for the $w_i$ is the same, and of course also $|f(x)-f(s_0)|=|x-s_0|$ and $|f(x)-f(s_k)|=|x-s_k|$, which implies that the corresponding computation for the coefficients $b_i$ produces the same result. We conclude $f(x)-f(s_0)=\sum a_iw_i=\phi(\sum a_i u_i)=\phi(x-s_0)$. $\square$

Now let $U^\perp, W^\perp$ be the orthogonal complements of $U,W$. From $\dim U=\dim W$ we have $\dim U^\perp=\dim W^\perp$, hence there esists an isometric linear isomorphism $\psi\colon U^\perp \to W^\perp$. Let $A\colon \Bbb R^n=U\oplus U^\perp \to \Bbb R^n=W\oplus W^\perp$ be the linear map $u+u^\perp \mapsto \phi(u)+\psi(u^\perp)$. Then $A$ is an isomety, hence $|\det A|=1$. If we let $b=f(s_0)-As_0$ we obtain (using the lemma above) $$ f(x)=Ax+b\qquad\text{for all }x\in S.$$


If $U^\perp$ (and $W^\perp$) has positive dimension (or equivalently, if $U$ is a proper subspace) then we have several choices for the isometry $\psi$ (e.g., replace $\psi$ with $-\psi$). Hence uniqueness of $A$ and $b$ can hold only if $U=W=\Bbb R^n$. To check this, assume $U=W=\Bbb R^n$and that also $f(x)=A'x+b'$ for all $x\in S$ with some other matrix $A'$ and vector $b'$. Then $A'u_i=A'(s_i-s_0)=f(s_i)-f(s_0)=A(s_i-s_0)=Au_i$ for $1\le i\le m=n$. As the $u_i$ form a basis of $\Bbb R^n$, we conclude that $A'=A$. But then also $b'=f(s_0)-A's_0=f(s_0)-As_0=b$. We conclude that $A,b$ are uniquely determined if and only if $S$ is not contained in a proper affine subspace.

  • 2
    I do not understand why the $w_i$ form a basis for $W$? Do you assume $f$ is affine? Thanks. – Asaf Shachar Apr 03 '17 at 15:45
  • @AsafShachar, It can be proved without assuming $f$ is affine, but it requires some work. Today I managed to fill all the details and soon I'll post my own answer. – Kulisty Nov 05 '20 at 21:49
  • You can prove this in the following way: consider map p: S-s0 -> f(S)-f(s0). It maps 0 to 0 and ui to wi. It is also an isometry. You can prove, that if you choose n vectors from S-s0, then its Gram matrix will be equal to Gram matrix of images of these n vectors. You need the lemma: Gram matrix of n vectors is invertible if and only if these vectors are linearly independent. So, since ui are linearly independent, wi are also(Gram matrix stays invertible). We prove that wi generate W by contradiction: if wi doesn't generqte W, then there are n+1 linearly independent vectors in W, but – kostya2139 Nov 17 '24 at 01:09
  • Since Gram matrix stays, that means that there are n+1 linearly independent vectors in S-s0 - contradiction.(We are using, that W is generated by p(S-s0)) – kostya2139 Nov 17 '24 at 01:17